U of W models when states can ease social distancing

[REWahoo]

When Is It Safe To Ease Social Distancing? Here's What One Model Says For Each State

The consensus view is that states shouldn't open up unless they have a robust system to detect and quash new flare-ups by testing to see who is infected, tracing their contacts, and isolating and quarantining as needed.

It appears that isn't going to happen, at least not within the next several months. And since that's the case...

...one team of disease modelers — from the University of Washington's Institute for Health Metrics and Evaluation or IHME — ...are asking the question: What is the maximum number of new infections that states could handle with their current testing and contact tracing capacity?

IHME's answer: 1 new infection per million people in a given state. They estimate that states with this level of transmission should be able to keep outbreaks from flaring up even after people start mingling again, though the researchers stress that states would still need to limit large gatherings.

IHME's team built a model to forecast when each state will reach that threshold of 1 new infection per million. Their main finding is that very few states are close.

The estimated date each state will reach that 1 new infection per million level ranges from May 6 (Hawaii, Montana) to July 19 (North Dakota). A few highlights:

Ohio - May 14
California - May 18
New Jersey, New York, Oregon, Pennsylvania - May 22
South Carolina, Texas - June 8
Florida, Kentucky - June 14
Arkansas, Georgia - June 22

 

[martyp]

The problem with the U of W model is that they report by state and not by city or region. Hospital resources are not evenly distributed and it matters how the infections are distributed. They have the data. It shouldn't be too hard to break it down to regions that would give a little more usable information. For example, it doesn't matter what medical resources are available in Seattle if there is an outbreak in Spokane

 

[timo2]

Have you looked at the Los Alamos National Lab model? It is a little different. The drill down for each State is interesting.

COVID-19 Confirmed and Forecasted Case Data

https://covid-19.bsvgateway.org/


It turns out the LANL won last year's CDC flu forecasting contest.
https://www.lanl.gov/discover/news-release-archive/2019/October/1022-flu-forecasting-challenge.php

 

[CaptTom]

1 new infection per million people in a given state.

But, how would you ever know that without testing? I find current case count numbers to be totally meaningless. No-one can even agree what percentage of people have already had an infection, often asymptomatically.

 

[REWahoo]

But, how would you ever know that without testing? I find current case count numbers to be totally meaningless. No-one can even agree what percentage of people have already had an infection, often asymptomatically.

As I understand it, the model assumes states continue their current level of testing, however flawed or inadequate that may be.

What is the maximum number of new infections that states could handle with their current testing and contact tracing capacity?

 

[Aerides]

The testing is showing an uptick the past few days at the national level. So, yes, continued increase on that would help.

But I agree the state level thing... I mean Florida feels like 2 or 3 different states on a non-covid day anyway. Fortunately, at least here, a lot of decisions are being made at the county and city level.

Boca Raton closed the city beaches a week before Palm Beach county closed the other few. The state followed another week after that at least. I expect similar spacing as we open back up.

 

[Montecfo]

These models have been so spectacularly wrong, not sure how much confidence we should have in any conclusions suggested.

 

[Chuckanut]

These models have been so spectacularly wrong, not sure how much confidence we should have in any conclusions suggested.

I don't think it's the models being wrong but rather people don't understand their many limitations thus they are misused and misrepresented.

Models are a great starting point when tackling new problems. They are far better than just guessing and making random assumptions. They focus the experts. OTOH, models for predicting weather, a very old problem, have proven to be very accurate. I know because I vary rarely hear jokes about the weatherman these days. He/She is just to good!

 

[Montecfo]

I don't think it's the models being wrong but rather people don't understand their many limitations thus they are misused and misrepresented.

Models are a great starting point when tackling new problems. They are far better than just guessing and making random assumptions. They focus the experts. OTOH, models for predicting weather, a very old problem, have proven to be very accurate. I know because I vary rarely hear jokes about the weatherman these days. He/She is just to good!

Oh I am a fan of modelling in general. Coronavirus modeling seems to have been worse than a guess when they are so far off.

Models can suggest precision, but that can be an illusion, as we have seen.

 

[stepford]

The problem isn't so much the models as the way in which their results have been presented and interpreted to the public by various talking heads. The model may say x fatalities +- y under certain assumptions. Then health care professionals and the politicians that listen to them warn the public using the possible number x+y while others trying to justify a more lax approach may talk of x-y. Meanwhile nobody talks about the assumptions going into the model.

Just eyeballing things right now as we zero in on 55-60,000 fatalities in the US by the end of April it seems that 60,000 - 240,000 eventual fatality estimate from a few weeks back doesn't seem that far off (within the error bar as we used to say in the research biz). More recent models have tightened up those error bounds as more and better data is available, but that doesn't mean the earlier models were intrinsically flawed, simply that they had limited information and very broad error bounds as a result.

 

[ivinsfan]

The problem with the U of W model is that they report by state and not by city or region. Hospital resources are not evenly distributed and it matters how the infections are distributed. They have the data. It shouldn't be too hard to break it down to regions that would give a little more usable information. For example, it doesn't matter what medical resources are available in Seattle if there is an outbreak in Spokane

Our state has two issues right now, nursing homes infections and issues with a couple of meat packing plants...so do you lock everyone down until both these problems are fixed or adapt and had different rules for different things.

For example what exacting do they mean when they say "social distancing" relaxing, just too many uncertainties.

Half the time I feel like these institutions want their ten minutes of fame.

 

[ivinsfan]

The problem isn't so much the models as the way in which their results have been presented and interpreted to the public by various talking heads. The model may say x fatalities +- y under certain assumptions. Then health care professionals and the politicians that listen to them warn the public using the possible number x+y while others trying to justify a more lax approach may talk of x-y. Meanwhile nobody talks about the assumptions going into the model.

Just eyeballing things right now as we zero in on 55-60,000 fatalities in the US by the end of April it seems that 60,000 - 240,000 eventual fatality estimate from a few weeks back doesn't seem that far off (within the error bar as we used to say in the research biz). More recent models have tightened up those error bounds as more and better data is available, but that doesn't mean the earlier models were intrinsically flawed, simply that they had limited information and very broad error bounds as a result.

It is the models because they all tweak them one way or another and just let the results out to the world.

 

[OldShooter]

These models have been so spectacularly wrong, not sure how much confidence we should have in any conclusions suggested.

Humankind has never created a model or a method that produces useful forecasts for any complex system. That will be true here, has been true for economic and stock market models, and will be shown to be true for climate forecasting. There is no reason to expect that this situation will be any different.

That said, models can be educational in trying to understand the consequences of tweeking parameters -- sensitivity analysis. But even that assumes that the modeler knows what the input parameters are. Easy example yesterday: Now they are saying that they don't know if people develop immunity after they have the disease.

It's basically the chaos theory problem. The butterfly in India flutters and causes a hurricane in the South Atlantic.

 

[ivinsfan]

When Is It Safe To Ease Social Distancing? Here's What One Model Says For Each State



It appears that isn't going to happen, at least not within the next several months. And since that's the case...



The estimated date each state will reach that 1 new infection per million level ranges from May 6 (Hawaii, Montana) to July 19 (North Dakota). A few highlights:

Ohio - May 14
California - May 18
New Jersey, New York, Oregon, Pennsylvania - May 22
South Carolina, Texas - June 8
Florida, Kentucky - June 14
Arkansas, Georgia - June 22

This one new infection for every million is a pipe dream , not going to happen and unrealistic we sure have medical care for more then one in one million.

 

[donheff]

The problem isn't so much the models as the way in which their results have been presented and interpreted to the public by various talking heads. The model may say x fatalities +- y under certain assumptions. Then health care professionals and the politicians that listen to them warn the public using the possible number x+y while others trying to justify a more lax approach may talk of x-y. Meanwhile nobody talks about the assumptions going into the model.

Just eyeballing things right now as we zero in on 55-60,000 fatalities in the US by the end of April it seems that 60,000 - 240,000 eventual fatality estimate from a few weeks back doesn't seem that far off (within the error bar as we used to say in the research biz). More recent models have tightened up those error bounds as more and better data is available, but that doesn't mean the earlier models were intrinsically flawed, simply that they had limited information and very broad error bounds as a result.

I listened to a podcast with a lead scientist at IHME being interviewed by Nate Silver of 538.org. The IHME researcher pointed out that various estimates by other people have been conflated with the IHME estimates confusing the issue. For example they never had 250k in their range. That was someone else’s wag. IHME’s first range was something close to 35k - 135k (I didn’t jot it down but that’s close). They have since narrowed the range as data have come in.

 

[cathy63]

The IHME model actually does model some states by area where there is adequate data to do so. They're located in Washington, so that may be the reason they've split their models for that state into several sub-models. They also have regional models for a lot of European countries now.

For those like me, who are interested in what's going into the models, and how those who create them think about their work, I recommended this podcast in another thread a few days ago: https://fivethirtyeight.com/videos/how-one-modeler-is-trying-to-forecast-the-toll-of-covid-19/

Thanks for the pointer to the Los Alamos model.

 

[audreyh1]

Just eyeballing things right now as we zero in on 55-60,000 fatalities in the US by the end of April it seems that 60,000 - 240,000 eventual fatality estimate from a few weeks back doesn't seem that far off (within the error bar as we used to say in the research biz). More recent models have tightened up those error bounds as more and better data is available, but that doesn't mean the earlier models were intrinsically flawed, simply that they had limited information and very broad error bounds as a result.

We have already crossed 55,000 deaths in the US, and numbers seem to drop over the weekend and then catch back up early in the week. Yep, I expect we’ll cross that 60,000 number by the end of the month.

One model mentioned somewhere was projecting 65K US deaths by early August? Oops.

For the record, the current modeling projects 779 total deaths in Washington state due to COVID-19 by Aug. 4, compared with 23,741 deaths in New York and 65,976 deaths in the nation. Johns Hopkins University’s coronavirus tracker currently reports 824,147 confirmed COVID-19 cases and 45,039 deaths in the United States. Over the past few days, the cumulative death toll in North America has been growing by roughly 2,000 daily.

IHME model. Article April 21. https://news.yahoo.com/coronavirus-computer-modelers-revise-dates-021424815.html

 

[ivinsfan]

We have already crossed 55,000 deaths in the US, and numbers seem to drop over the weekend and then catch back up early in the week. Yep, I expect we’ll cross that 60,000 number by the end of the month.

One model mentioned somewhere was projecting 65K US deaths by early August? Oops.

In New York they have said that deaths per day are a lagging indicator and most of the attention is on new cases per day as less cases per day should result in less deaths per day down the road..

In our state they also track number recovered ( either out of hospital or past the 14 day isolation period vs active cases..

I'm not sure anyone anticipated how difficult it would be to keep this out of nursing home and group living situation. Perhaps duee to the number of asymptomatic cases that seem to be around.

 

[MichaelB]

We have already crossed 55,000 deaths in the US, and numbers seem to drop over the weekend and then catch back up early in the week. Yep, I expect we’ll cross that 60,000 number by the end of the month.

One model mentioned somewhere was projecting 65K US deaths by early August? Oops.
IHME model. Article April 21. https://news.yahoo.com/coronavirus-computer-modelers-revise-dates-021424815.html

We’ve probably already crossed that unfortunate threshold, some deaths may have been misclassified, especially early on.

 

[audreyh1]

I'm not sure anyone anticipated how difficult it would be to keep this out of nursing home and group living situation. Perhaps due to the number of asymptomatic cases that seem to be around.

Yes, I think even the CDC was wishing and hoping that the asymptomatic spread was very minor, despite pretty strong indications otherwise reported from other countries and ships.

Elder care homes are super vulnerable and we don’t have the resources to turn them into isolation prisons detaining employees as well.

 

[donheff]

We have already crossed 55,000 deaths in the US, and numbers seem to drop over the weekend and then catch back up early in the week. Yep, I expect we’ll cross that 60,000 number by the end of the month.

One model mentioned somewhere was projecting 65K US deaths by early August? Oops.
IHME model. Article April 21. https://news.yahoo.com/coronavirus-computer-modelers-revise-dates-021424815.html

It is IHME that was projecting 65K and is now projecting 67,641 deaths by August (with an upper range of about 120K). They base it on states continuing to follow current social distancing until infections are minimized and containment achieved. They have data from other countries showing the backside slopes are steep.

Still, it is hard to believe the actually numbers won't be well above 67K.

 

[audreyh1]

Still, it is hard to believe the actually numbers won't be well above 67K.

Yes, since we should hit 67K deaths by very early May, and that’s with the restrictions in place until just a few days ago.

 

[MichaelB]

Here’s another informative interview with Scott Gottlieb, including a discussion on Remdesivir. StatNews podcast 109 https://www.statnews.com/category/readout-loud/

He is somewhsy positive on the drug, but leaves the impression it will not have a major impact. StatNews is a highly qualified news source that specializes in heath care. Usually behind a paywall, it’s COVID coverage is free.

 

[Sunset]

It is IHME that was projecting 65K and is now projecting 67,641 deaths by August (with an upper range of about 120K). They base it on states continuing to follow current social distancing until infections are minimized and containment achieved. They have data from other countries showing the backside slopes are steep.

Still, it is hard to believe the actually numbers won't be well above 67K.

May 2, and we are at 67,444

I saw some modeler of the virus claim that now they have increased the death rate to 72,000 by Aug...

I'm thinking we will hit that in about a week.

 

[Dtail]

May 2, and we are at 67,444

I saw some modeler of the virus claim that now they have increased the death rate to 72,000 by Aug...

I'm thinking we will hit that in about a week.

The IHME model appears to be very optimistic. May 2nd had ~ 1k deaths not including NY/NJ.
This model just keeps updating the projection a little at a time, as the actuals surpass the estimates.